feat(*): 生图与视频工作台 - 新增独立生成流程并整合平台配置

2026-10-13 09:04:34 +08:00
1 parent 4c1d2b4012
commit 1f3f2c6020
@@ -0,0 +1,393 @@
+package workbench
+
+import (
+	"encoding/json"
+	"errors"
+	"math"
+	"strconv"
+	"strings"
+
+	"yingce/backend/internal/kernel"
+	"yingce/backend/internal/model"
+)
+
+type TaskInput struct {
+	Kind           string          `json:"kind"`
+	ToolID         string          `json:"toolId"`
+	PresetID       string          `json:"presetId,omitempty"`
+	Workbench      *TaskWorkbench  `json:"workbench,omitempty"`
+	Prompt         string          `json:"prompt"`
+	Model          string          `json:"model,omitempty"`
+	LogicalModelID string          `json:"logicalModelId,omitempty"`
+	ChannelID      string          `json:"channelId,omitempty"`
+	ModelChoice    *ModelChoice    `json:"modelChoice,omitempty"`
+	Parameters     map[string]any  `json:"parameters,omitempty"`
+	References     []TaskReference `json:"references,omitempty"`
+}
+
+type TaskWorkbench struct {
+	Kind     string `json:"kind"`
+	ToolID   string `json:"toolId"`
+	PresetID string `json:"presetId,omitempty"`
+}
+
+type ModelChoice struct {
+	LogicalModelID string `json:"logicalModelId,omitempty"`
+	UserChannelID  string `json:"userChannelId,omitempty"`
+	ModelName      string `json:"modelName,omitempty"`
+}
+
+type TaskReference struct {
+	ResourceID string `json:"resourceId"`
+	Kind       string `json:"kind"`
+	Role       string `json:"role,omitempty"`
+	SlotKey    string `json:"slotKey,omitempty"`
+}
+
+type TaskResolution struct {
+	TaskType       string
+	Operation      string
+	Input          map[string]any
+	Parameters     map[string]any
+	Prompt         string
+	Model          string
+	LogicalModelID string
+	ChannelID      string
+	Tool           *model.WorkbenchTool
+	Preset         *model.WorkbenchPreset
+}
+
+func (s *Service) ResolveTask(userID string, req TaskInput) (*TaskResolution, error) {
+	toolID := strings.TrimSpace(req.ToolID)
+	presetID := strings.TrimSpace(req.PresetID)
+	if req.Workbench != nil && (req.Workbench.Kind != req.Kind || strings.TrimSpace(req.Workbench.ToolID) != toolID || (strings.TrimSpace(req.Workbench.PresetID) != "" && strings.TrimSpace(req.Workbench.PresetID) != presetID)) {
+		return nil, kernel.BadAuthRequest("工作台任务范围字段不一致")
+	}
+	if !validKind(req.Kind) {
+		return nil, kernel.BadAuthRequest("工作台类型无效")
+	}
+	if toolID == "" && presetID != "" {
+		return nil, kernel.BadAuthRequest("使用提示词预设前必须选择创作工具")
+	}
+	var tool *model.WorkbenchTool
+	var preset *model.WorkbenchPreset
+	if toolID != "" {
+		var err error
+		tool, preset, err = s.repo.WorkbenchToolAndPreset(toolID, presetID, true)
+		if err != nil {
+			return nil, kernel.BadAuthRequest("工具或预设未开放")
+		}
+		if tool.Kind != req.Kind {
+			return nil, kernel.BadAuthRequest("所选工具与工作台类型不匹配")
+		}
+	}
+	if strings.TrimSpace(req.Prompt) == "" && (tool == nil || toolRequiresPrompt(tool)) {
+		return nil, kernel.BadAuthRequest("当前工具需要填写提示词")
+	}
+	modelName, logicalModelID, channelID, providerConfig, err := resolveModelInput(req)
+	if err != nil {
+		return nil, err
+	}
+	if len(providerConfig) == 0 {
+		return nil, kernel.BadAuthRequest("必须提供可执行模型配置")
+	}
+	if containsSensitiveURL(providerConfig) {
+		return nil, kernel.BadAuthRequest("浏览器不能直接提交模型配置、URL 或凭据")
+	}
+	if len(req.References) > 20 {
+		return nil, kernel.BadAuthRequest("参考素材数量超出限制")
+	}
+	parameters, err := normalizeTaskParameters(req.Kind, req.Parameters)
+	if err != nil {
+		return nil, err
+	}
+	refs := make([]TaskReference, 0, len(req.References))
+	frameRoles := map[string]bool{}
+	for _, ref := range req.References {
+		if strings.TrimSpace(ref.ResourceID) == "" || !validReferenceKind(ref.Kind) {
+			return nil, kernel.BadAuthRequest("参考素材无效")
+		}
+		role := strings.TrimSpace(ref.Role)
+		if !validReferenceRole(req.Kind, ref.Kind, role) {
+			return nil, kernel.BadAuthRequest("参考素材角色无效")
+		}
+		if role == "first_frame" || role == "last_frame" {
+			if frameRoles[role] {
+				return nil, kernel.BadAuthRequest("首帧和尾帧各只能指定一个素材")
+			}
+			frameRoles[role] = true
+		}
+		resource, err := s.repo.ResourceForUser(userID, strings.TrimSpace(ref.ResourceID))
+		if err != nil || resource.Status != model.ResourceStatusReady || resource.Kind != ref.Kind {
+			return nil, kernel.BadAuthRequest("参考素材不存在、未就绪或类型不匹配")
+		}
+		refs = append(refs, TaskReference{ResourceID: resource.ID, Kind: resource.Kind, Role: role, SlotKey: strings.TrimSpace(ref.SlotKey)})
+	}
+	if tool != nil {
+		capability := map[string]any{}
+		if err := json.Unmarshal([]byte(tool.CapabilityJSON), &capability); err != nil {
+			return nil, errors.New("工具能力配置无效")
+		}
+		if err := validateTaskCapabilities(tool.Kind, capability, refs); err != nil {
+			return nil, err
+		}
+	}
+	metadata := map[string]any{"source": "workbench", "workbenchKind": req.Kind}
+	workbenchSnapshot := map[string]any{"kind": req.Kind, "parameters": parameters}
+	snapshotReferences := make([]map[string]any, 0, len(refs))
+	for _, ref := range refs {
+		reference := map[string]any{"resourceId": ref.ResourceID, "kind": ref.Kind}
+		if ref.Role != "" {
+			reference["role"] = ref.Role
+		}
+		if ref.SlotKey != "" {
+			reference["slotKey"] = ref.SlotKey
+		}
+		snapshotReferences = append(snapshotReferences, reference)
+	}
+	workbenchSnapshot["references"] = snapshotReferences
+	if tool != nil {
+		metadata["workbenchToolId"], metadata["workbenchToolName"], metadata["workbenchToolVersion"] = tool.ID, tool.Name, tool.Version
+		workbenchSnapshot["toolId"], workbenchSnapshot["toolName"], workbenchSnapshot["toolVersion"] = tool.ID, tool.Name, tool.Version
+	}
+	metadata["workbench"] = workbenchSnapshot
+	input := map[string]any{"mode": req.Kind, "prompt": strings.TrimSpace(req.Prompt), "config": providerConfig, "metadata": metadata}
+	if len(refs) > 0 {
+		imageRefs, videoRefs, audioRefs := []map[string]any{}, []map[string]any{}, []map[string]any{}
+		for _, ref := range refs {
+			value := map[string]any{"resourceId": ref.ResourceID, "storageKey": "resource:" + ref.ResourceID}
+			if ref.Role != "" {
+				value["role"] = ref.Role
+			}
+			if ref.SlotKey != "" {
+				value["slotKey"] = ref.SlotKey
+			}
+			switch ref.Kind {
+			case "image":
+				imageRefs = append(imageRefs, value)
+			case "video":
+				videoRefs = append(videoRefs, value)
+			case "audio":
+				audioRefs = append(audioRefs, value)
+			}
+		}
+		if len(imageRefs) > 0 {
+			input["referenceImages"] = imageRefs
+		}
+		if len(videoRefs) > 0 {
+			input["referenceVideos"] = videoRefs
+		}
+		if len(audioRefs) > 0 {
+			input["referenceAudios"] = audioRefs
+		}
+	}
+	if preset != nil {
+		meta := input["metadata"].(map[string]any)
+		meta["workbenchPresetId"], meta["workbenchPresetVersion"], meta["workbenchPresetName"] = preset.ID, preset.Version, preset.Name
+		meta["workbenchPresetPrompt"] = preset.PromptTemplate
+		workbenchSnapshot := meta["workbench"].(map[string]any)
+		workbenchSnapshot["presetId"], workbenchSnapshot["presetName"], workbenchSnapshot["presetVersion"] = preset.ID, preset.Name, preset.Version
+		workbenchSnapshot["presetPrompt"] = preset.PromptTemplate
+	}
+	return &TaskResolution{TaskType: map[string]string{KindImage: "canvas_image", KindVideo: "canvas_video"}[req.Kind], Operation: map[string]string{KindImage: "image", KindVideo: "video"}[req.Kind], Input: input, Parameters: parameters, Prompt: strings.TrimSpace(req.Prompt), Model: modelName, LogicalModelID: logicalModelID, ChannelID: channelID, Tool: tool, Preset: preset}, nil
+}
+
+func normalizeTaskParameters(kind string, params map[string]any) (map[string]any, error) {
+	allowed := map[string]map[string]bool{
+		KindImage: {"size": true, "quality": true, "count": true},
+		KindVideo: {"seconds": true, "ratio": true, "resolution": true, "generateAudio": true, "watermark": true},
+	}[kind]
+	if allowed == nil {
+		return nil, kernel.BadAuthRequest("工作台类型无效")
+	}
+	normalized := make(map[string]any, len(params))
+	for key, value := range params {
+		if !allowed[key] {
+			return nil, kernel.BadAuthRequest("生成参数包含不支持字段: " + key)
+		}
+		switch key {
+		case "size", "quality", "ratio", "resolution":
+			text, ok := value.(string)
+			text = strings.TrimSpace(text)
+			if !ok || len(text) > 64 || strings.Contains(text, "://") || strings.HasPrefix(strings.ToLower(text), "data:") {
+				return nil, kernel.BadAuthRequest("生成参数格式无效: " + key)
+			}
+			if text != "" {
+				normalized[key] = text
+			}
+		case "count", "seconds":
+			number, ok := taskParameterInteger(value)
+			max := 20
+			if key == "seconds" {
+				max = 3600
+			}
+			if !ok || number < 1 || number > max {
+				return nil, kernel.BadAuthRequest("生成参数数值无效: " + key)
+			}
+			normalized[key] = number
+		case "generateAudio", "watermark":
+			flag, ok := value.(bool)
+			if !ok {
+				return nil, kernel.BadAuthRequest("生成参数格式无效: " + key)
+			}
+			normalized[key] = flag
+		}
+	}
+	return normalized, nil
+}
+
+func taskParameterInteger(value any) (int, bool) {
+	switch number := value.(type) {
+	case int:
+		return number, true
+	case int64:
+		return int(number), int64(int(number)) == number
+	case float64:
+		if math.IsNaN(number) || math.IsInf(number, 0) || math.Trunc(number) != number {
+			return 0, false
+		}
+		return int(number), float64(int(number)) == number
+	case json.Number:
+		parsed, err := strconv.Atoi(number.String())
+		return parsed, err == nil
+	default:
+		return 0, false
+	}
+}
+
+func validReferenceRole(kind, referenceKind, role string) bool {
+	if role == "" || role == "reference" {
+		return kind == KindVideo || referenceKind == "image"
+	}
+	return kind == KindVideo && referenceKind == "image" && (role == "first_frame" || role == "last_frame")
+}
+
+func resolveModelInput(req TaskInput) (string, string, string, map[string]any, error) {
+	modelName := strings.TrimSpace(req.Model)
+	logicalModelID := strings.TrimSpace(req.LogicalModelID)
+	channelID := strings.TrimSpace(req.ChannelID)
+	if req.ModelChoice != nil {
+		choice := req.ModelChoice
+		if choice.UserChannelID != "" {
+			return "", "", "", nil, kernel.BadAuthRequest("工作台暂不接受浏览器自定义渠道选择")
+		}
+		if value := strings.TrimSpace(choice.LogicalModelID); value != "" {
+			if logicalModelID != "" && logicalModelID != value {
+				return "", "", "", nil, kernel.BadAuthRequest("模型选择字段不一致")
+			}
+			logicalModelID = value
+		}
+		if value := strings.TrimSpace(choice.ModelName); value != "" {
+			if modelName != "" && modelName != value {
+				return "", "", "", nil, kernel.BadAuthRequest("模型名称字段不一致")
+			}
+			modelName = value
+		}
+	}
+	if modelName == "" {
+		return "", "", "", nil, kernel.BadAuthRequest("必须选择一个模型")
+	}
+	// 创作页的系统目录模型多数只有渠道 ID + 模型名，没有 logicalModelId。
+	// 两者都空才拒绝；渠道是否为系统渠道由后续 admission 再校验。
+	if logicalModelID == "" && channelID == "" {
+		return "", "", "", nil, kernel.BadAuthRequest("工作台任务必须选择系统模型目录中的模型")
+	}
+	config := map[string]any{"mode": req.Kind, "model": modelName}
+	if channelID != "" {
+		config["channelId"] = channelID
+	}
+	return modelName, logicalModelID, channelID, config, nil
+}
+
+func validReferenceKind(kind string) bool {
+	return kind == "image" || kind == "video" || kind == "audio"
+}
+
+func containsSensitiveURL(value any) bool {
+	switch item := value.(type) {
+	case map[string]any:
+		for key, child := range item {
+			lower := strings.ToLower(key)
+			if lower == "api_key" || lower == "apikey" || lower == "authorization" || lower == "headers" || lower == "baseurl" || lower == "base_url" || lower == "url" {
+				return true
+			}
+			if containsSensitiveURL(child) {
+				return true
+			}
+		}
+	case []any:
+		for _, child := range item {
+			if containsSensitiveURL(child) {
+				return true
+			}
+		}
+	}
+	return false
+}
+
+func validateTaskCapabilities(kind string, capability map[string]any, refs []TaskReference) error {
+	allowedRefs, _ := capability["referenceKinds"].([]any)
+	allowedSet := map[string]bool{}
+	for _, value := range allowedRefs {
+		if text, ok := value.(string); ok {
+			allowedSet[text] = true
+		}
+	}
+	var contract InputContract
+	if raw, ok := capability["inputContract"]; ok {
+		encoded, _ := json.Marshal(raw)
+		_ = json.Unmarshal(encoded, &contract)
+	}
+	slots := map[string]ReferenceSlot{}
+	for _, slot := range contract.Slots {
+		slots[slot.Key] = slot
+	}
+	counts := map[string]int{}
+	for _, ref := range refs {
+		if len(allowedSet) > 0 && !allowedSet[ref.Kind] {
+			return kernel.BadAuthRequest("当前工具不支持该类型参考素材")
+		}
+		if kind == KindImage && ref.Kind != "image" {
+			return kernel.BadAuthRequest("生图任务只允许使用图片参考素材")
+		}
+		if kind == KindVideo && ref.Kind != "image" && ref.Kind != "video" && ref.Kind != "audio" {
+			return kernel.BadAuthRequest("视频任务参考素材类型无效")
+		}
+		if len(contract.Slots) == 0 {
+			continue
+		}
+		slotKey := strings.TrimSpace(ref.SlotKey)
+		slot, ok := slots[slotKey]
+		if !ok {
+			return kernel.BadAuthRequest("参考素材未指定有效的输入槽位")
+		}
+		mediaAllowed := false
+		for _, mediaType := range slot.MediaTypes {
+			if mediaType == ref.Kind {
+				mediaAllowed = true
+				break
+			}
+		}
+		if !mediaAllowed {
+			return kernel.BadAuthRequest("参考素材类型与输入槽位不匹配")
+		}
+		counts[slotKey]++
+		if counts[slotKey] > slot.MaxCount {
+			return kernel.BadAuthRequest("输入槽位的参考素材数量超出上限")
+		}
+	}
+	for _, slot := range contract.Slots {
+		if counts[slot.Key] < slot.MinCount || (slot.Required && counts[slot.Key] == 0) {
+			return kernel.BadAuthRequest("请补齐必需的输入素材：" + slot.Name)
+		}
+	}
+	return nil
+}
+
+func toolRequiresPrompt(tool *model.WorkbenchTool) bool {
+	if tool == nil {
+		return true
+	}
+	config, err := parseCatalogConfig(tool.CapabilityJSON, tool.GuideJSON)
+	return err != nil || config.InputContract.PromptRequired || config.InputContract.MinTextReferences > 0
+}